PR 18313
bin * ieee.c (ieee_read_cxx_class): Initialise the varargs variable.
* readelf.c (uncompress_section_contents): Zero initialise the
zstream structure.
bfd * compress.c (decompress_contents): Zero initialse the z_stream
structure.
Bits in section flags should only be defined in bfd/section.c. This
patch moves SEC_MEP_VLIW to bfd/section.c.
bfd/
PR binutils/18316
* section.c (SEC_MEP_VLIW): New.
* bfd-in2.h: Regenerated.
include/elf/
PR binutils/18316
* mep.h (SEC_MEP_VLIW): Removed.
When we set up st_name for output section name in elf_fake_sections, we
don't know if the compressed DWARF debug section will be smaller. We may
end up with compressed DWARF debug sections which are bigger than the
uncompressed ones. This patch delays setting up st_name for output DWARF
debug section to _bfd_elf_assign_file_positions_for_non_load which will
compress the output debug section. We also postpone placement of shstrtab
section after DWARF debug sections have been compressed. The net effect
is .shstrtab section is now placed after .symtab and .strtab sections.
bfd/
PR ld/18277
* compress.c (bfd_compress_section_contents): Remove the
write_compress argument.
(bfd_init_section_compress_status): Updated.
(bfd_compress_section): Likewise.
* elf.c (_bfd_elf_set_reloc_sh_name): New.
(_bfd_elf_init_reloc_shdr): Add delay_st_name_p. Set sh_name
to (unsigned int) -1 if delay_st_name_p is TRUE. Use
_bfd_elf_set_reloc_sh_name.
(elf_fake_sections): Don't rename DWARF debug section for
linker output if it will be compressed. Instead, set
delay_st_name_p to TRUE and pass it to _bfd_elf_init_reloc_shdr.
(assign_section_numbers): Call _bfd_elf_strtab_addref only if
sh_name != (unsigned int) -1. Don't finalize nor assign
shstrtab section here. Delay setting output section names to
_bfd_elf_write_object_contents.
(_bfd_elf_compute_section_file_positions): Update comments on
sh_offset for shstrtab section.
(assign_file_positions_for_non_load_sections): Set sh_offset to
-1 for shstrtab section.
(assign_file_positions_except_relocs): Likewise.
(_bfd_elf_assign_file_positions_for_non_load): Set up sh_name
when compressing DWARF debug sections. Place shstrtab section
after DWARF debug sections have been compressed.
(_bfd_elf_write_object_contents): Setting sh_name for output
sections.
ld/testsuite/
PR ld/18277
* ld-elf/compressed1d.d: New.
* ld-elf/compressed1e.d: Likewise.
Change compressed input debug section name for objdump is very confusing.
But we need to change it for linker so that linker will consider the
input section as a debug section. This patch delays section rename to
elf_fake_sections for objcopy and avoids it for objdump.
bfd/
PR binutils/18209
* bfd.c (bfd): Add is_linker_input.
* elf.c (convert_debug_to_zdebug): New.
(convert_zdebug_to_debug): Likewise.
(_bfd_elf_make_section_from_shdr): Don't convert .debug_* to
.zdebug_* here. Use convert_zdebug_to_debug. Set SEC_ELF_RENAME.
(_bfd_elf_init_reloc_shdr): Pass a pointer to section name
instead of a pointer to section.
(elf_fake_sections): Rename the section name if SEC_ELF_RENAME
is set.
* section.c (SEC_ELF_RENAME): New.
* bfd-in2.h: Regenerated.
binutils/
PR binutils/18209
* objcopy.c (setup_section): Copy compress status.
binutils/testsuite/
PR binutils/18209
* binutils-all/compress.exp: Replace dw2-3.W with dw2-3gabi.W
on zlib-gabi output.
* binutils-all/dw2-1.W: Convert section names to .zdebug_*.
* binutils-all/dw2-3.W: Likewise.
* binutils-all/objdump.W: Likewise.
* binutils-all/dw2-3gabi.W: New file.
ld/
PR binutils/18209
* ldfile.c (ldfile_try_open_bfd): Set is_linker_input to 1.
commit 9ad5cbcfb2 added the initial support
for more than 64k ELF sections with holes for reserved section indices in
section header table and set entries of reserved section indices in
section header table to index 0:
for (secn = 1; secn < section_number; ++secn)
- i_shdrp[secn]->sh_name = _bfd_elf_strtab_offset (elf_shstrtab (abfd),
- i_shdrp[secn]->sh_name);
-
+ if (i_shdrp[secn] == NULL)
+ i_shdrp[secn] = i_shdrp[0];
+ else
+ i_shdrp[secn]->sh_name = _bfd_elf_strtab_offset (elf_shstrtab (abfd),
+ i_shdrp[secn]->sh_name);
commit 4fbb74a605 removed holes in section
header table. Check for i_shdrp[secn] == NULL is no longer needed now.
This patch removes it.
* elf.c (assign_section_numbers): Always set up sh_name.
Iterate ELF sections by pointer without section counter. It removes
"i++".
* elf.c (assign_file_positions_for_non_load_sections): Iterate
sections by pointer.
(_bfd_elf_assign_file_positions_for_non_load): Likewise.
This patch adds --compress-debug-sections=[none|zlib|zlib-gnu|zlib-gabi]
to ld for ELF targets to support generating compressed DWARF debug
sections. We always generate .zdebug_* section since section names have
been finalized and they can't be changed easily when compression is
being performed.
bfd/
* bfd-in.h (compressed_debug_section_type): New.
* compress.c (bfd_compress_section_contents): Add an argument
for linker write compression and always generate .zdebug_*
section when linking.
(bfd_init_section_compress_status): Pass FALSE to
bfd_compress_section_contents.
(bfd_compress_section): New function.
* elf.c (elf_fake_sections): For linking, set SEC_ELF_COMPRESS
on DWARF debug sections if COMPRESS_DEBUG is set and rename
section if COMPRESS_DEBUG_GABI_ZLIB isn't set.
(assign_file_positions_for_non_load_sections): Set sh_offset
to -1 if SEC_ELF_COMPRESS is set.
(assign_file_positions_except_relocs): Likwise.
(_bfd_elf_assign_file_positions_for_relocs): Renamed to ...
(_bfd_elf_assign_file_positions_for_non_load): This. Change
return time to bfd_boolean. Compress the section if
SEC_ELF_COMPRESS is set.
(_bfd_elf_write_object_contents): Updated.
(_bfd_elf_set_section_contents): Write section contents to
the buffer if SEC_ELF_COMPRESS is set.
* merge.c: Include "elf-bfd.h".
(sec_merge_emit): Add arguments for contents and offset. Write
to contents with offset if contents isn't NULL.
(_bfd_write_merged_section): Write section contents to the
buffer if SEC_ELF_COMPRESS is set. Pass contents and
output_offset to sec_merge_emit.
* elflink.c (bfd_elf_final_link): Allocate the buffer for
output section contents if SEC_ELF_COMPRESS is set.
* section.c (SEC_ELF_COMPRESS): New.
* bfd-in2.h: Regenerated.
gas/
* as.h (compressed_debug_section_type): Removed.
include/
* bfdlink.h (bfd_link_info): Add compress_debug.
ld/
* ld.texinfo: Document --compress-debug-sections=.
* ldmain.c (main): Set BFD_COMPRESS on output_bfd if
COMPRESS_DEBUG is set. Set BFD_COMPRESS_GABI on output_bfd
for COMPRESS_DEBUG_GABI_ZLIB.
* lexsup.c (elf_static_list_options): Add
--compress-debug-sections=.
* emultempl/elf32.em (OPTION_COMPRESS_DEBUG): New.
(xtra_long): Add "compress-debug-sections".
(gld${EMULATION_NAME}_handle_option): Handle
OPTION_COMPRESS_DEBUG.
ld/testsuite/
* ld-elf/compress.exp (build_tests): Add tests for
--compress-debug-sections=.
(run_tests): Likewise.
Add additonal tests for --compress-debug-sections=.
* ld-elf/gabiend.rt: New file.
* ld-elf/gabinormal.rt: Likewise.
* ld-elf/gnubegin.rS: Likewise.
* ld-elf/gnunormal.rS: Likewise.
* ld-elf/zlibbegin.rS: Likewise.
* ld-elf/zlibnormal.rS: Likewise.
gas * config/tc-rl78.h (TC_LINKRELAX_FIXUP): Define.
(TC_FORCE_RELOCATION_SUB_SAME): Define.
(DWARF2_USE_FIXED_ADVANCE_PC): Define.
* gas/lns/lns.exp: Add RL78 to list of targets using
DW_LNS_fixed_advance_pc.
bfd * elf32-rl78.c (RL78_OP_REL): New macro.
(rl78_elf_howto_table): Use it for complex relocs.
(get_symbol_value): Handle the cases when the info or status
arguments are NULL.
(get_romstart): Cache the status returned by get_symbol_value.
(get_ramstart): Likewise.
(RL78_STACK_PUSH): Generate an error message if the stack
overflows.
(RL78_STACK_POP): Likewise for underflows.
(rl78_compute_complex_reloc): New function. Contains the basic
processing code for all RL78 complex relocs.
(rl78_special_reloc): New function. Provides special reloc
handling for complex relocs.
(rl78_elf_relocate_section): Use rl78_compute_complex_reloc.
(rl78_offset_for_reloc): Likewise.
binutils* readelf.c (target_specific_reloc_handling): Add code to handle
RL78 complex relocs.
Address of protected data defined in the shared library may be external,
i.e., due to copy relocation. By default, linker backend checks if
relocations against protected data symbols are valid for building shared
library and issues an error if relocation isn't allowed. The new option
override linker backend default. When -z noextern-protected-data is used,
updates on protected data symbols by another module won't be visibile
to the resulting shared library. This option is specific to ELF/i386
and ELF/x86-64.
bfd/
PR ld/pr17709
* elflink.c (_bfd_elf_adjust_dynamic_copy): Check
info->extern_protected_data when warning copy relocs against
protected symbols.
(_bfd_elf_symbol_refs_local_p): Check info->extern_protected_data
when checking protected non-function symbols.
include/
PR ld/pr17709
* bfdlink.h (bfd_link_info): Add extern_protected_data.
ld/
PR ld/pr17709
* ld.texinfo: Document "-z noextern-protected-data".
* ldmain.c (main): Initialize link_info.extern_protected_data
to -1.
* lexsup.c (elf_shlib_list_options): Add
"-z [no]extern-protected-data".
* emulparams/elf32_x86_64.sh: Source extern_protected_data.sh.
* emulparams/elf_i386.sh: Likewise.
* emulparams/elf_i386_be.sh: Likewise.
* emulparams/elf_i386_chaos.sh: Likewise.
* emulparams/elf_i386_ldso.sh: Likewise.
* emulparams/elf_i386_vxworks.sh: Likewise.
* emulparams/elf_k1om.sh: Likewise.
* emulparams/elf_l1om.sh: Likewise.
* emulparams/elf_x86_64.sh: Source extern_protected_data.sh.
(PARSE_AND_LIST_OPTIONS): Renamed to ...
(PARSE_AND_LIST_OPTIONS_BNDPLT): This.
(PARSE_AND_LIST_ARGS_CASE_Z): Renamed to ...
(PARSE_AND_LIST_ARGS_CASE_Z_BNDPLT): This.
(PARSE_AND_LIST_OPTIONS): Append $PARSE_AND_LIST_OPTIONS_BNDPLT.
(PARSE_AND_LIST_ARGS_CASE_Z): Append
$PARSE_AND_LIST_ARGS_CASE_Z_BNDPLT.
* emulparams/extern_protected_data.sh: New file.
ld/testsuite/
PR ld/pr17709
* ld-i386/i386.exp: Run protected6b.
* ld-i386/protected6b.d: New file.
* ld-x86-64/protected6b.d: Likewise.
* ld-x86-64/x86-64.exp: Run protected6b.
Recognize NT_X86_XSTATE notes in FreeBSD process cores. Recent
FreeBSD versions include a note containing the XSAVE state for each
thread in the process when XSAVE is in use. The note stores a copy of
the current XSAVE mask in a reserved section of the machine-defined
XSAVE state at the same offset as Linux's NT_X86_XSTATE note.
For native processes, use the PT_GETXSTATE_INFO ptrace request to
determine if XSAVE is enabled, and if so the active XSAVE state mask
(that is, the value of %xcr0 for the target process) as well as the
size of XSAVE state area. Use the PT_GETXSTATE and PT_SETXSTATE requests
to fetch and store the XSAVE state, respectively, in the BSD x86
native targets.
In addition, the FreeBSD amd64 and i386 native targets now include
"read_description" target methods to determine the correct x86 target
description for the current XSAVE mask. On FreeBSD amd64 this also
properly returns an i386 target description for 32-bit binaries which
allows the 64-bit GDB to run 32-bit binaries.
Note that the ptrace changes are in the BSD native targets, not the
FreeBSD-specific native targets since that is where the other ptrace
register accesses occur. Of the other BSDs, NetBSD and DragonFly use
XSAVE in the kernel but do not currently export the extended state via
ptrace(2). OpenBSD does not currently support XSAVE.
bfd/ChangeLog:
* elf.c (elfcore_grok_note): Recognize NT_X86_XSTATE on
FreeBSD.
(elfcore_write_xstatereg): Use correct note name on FreeBSD.
gdb/ChangeLog:
* amd64-tdep.c (amd64_target_description): New function.
* amd64-tdep.h: Export amd64_target_description and tdesc_amd64.
* amd64bsd-nat.c [PT_GETXSTATE_INFO]: New variable amd64bsd_xsave_len.
(amd64bsd_fetch_inferior_registers) [PT_GETXSTATE_INFO]: Handle
x86 extended save area.
(amd64bsd_store_inferior_registers) [PT_GETXSTATE_INFO]: Likewise.
* amd64bsd-nat.h: Export amd64bsd_xsave_len.
* amd64fbsd-nat.c (amd64fbsd_read_description): New function.
(_initialize_amd64fbsd_nat): Set "to_read_description" to
"amd64fbsd_read_description".
* amd64fbsd-tdep.c (amd64fbsd_core_read_description): New function.
(amd64fbsd_supply_xstateregset): New function.
(amd64fbsd_collect_xstateregset): New function.
Add "amd64fbsd_xstateregset".
(amd64fbsd_iterate_over_regset_sections): New function.
(amd64fbsd_init_abi): Set "xsave_xcr0_offset" to
"I386_FBSD_XSAVE_XCR0_OFFSET".
Add "iterate_over_regset_sections" gdbarch method.
Add "core_read_description" gdbarch method.
* i386-tdep.c (i386_target_description): New function.
* i386-tdep.h: Export i386_target_description and tdesc_i386.
* i386bsd-nat.c [PT_GETXSTATE_INFO]: New variable i386bsd_xsave_len.
(i386bsd_fetch_inferior_registers) [PT_GETXSTATE_INFO]: Handle
x86 extended save area.
(i386bsd_store_inferior_registers) [PT_GETXSTATE_INFO]: Likewise.
* i386bsd-nat.h: Export i386bsd_xsave_len.
* i386fbsd-nat.c (i386fbsd_read_description): New function.
(_initialize_i386fbsd_nat): Set "to_read_description" to
"i386fbsd_read_description".
* i386fbsd-tdep.c (i386fbsd_core_read_xcr0): New function.
(i386fbsd_core_read_description): New function.
(i386fbsd_supply_xstateregset): New function.
(i386fbsd_collect_xstateregset): New function.
Add "i386fbsd_xstateregset".
(i386fbsd_iterate_over_regset_sections): New function.
(i386fbsd4_init_abi): Set "xsave_xcr0_offset" to
"I386_FBSD_XSAVE_XCR0_OFFSET".
Add "iterate_over_regset_sections" gdbarch method.
Add "core_read_description" gdbarch method.
* i386fbsd-tdep.h: New file.
When ELF linker backend searchs the symbol table of an archive element,
it should skip the object which has been claimed by plugin.
PR ld/18250
* elflink.c (elf_link_is_defined_archive_symbol): Return FALSE
if the object has been claimed by plugin.
There is no need to check and initialize pagesize_m1 in cache_bmmap.
* cache.c (cache_bmmap): Move pagesize_m1 ... to
(pagesize_m1): Here.
(bfd_cache_init): Initialize pagesize_m1.
When checking R_386_GOTOFF/R_X86_64_GOTOFF64 for building shared library,
we should check SYMBOL_REFERENCES_LOCAL instead of SYMBOLIC_BIND to cover
more cases.
bfd/
* elf32-i386.c (elf_i386_relocate_section): Replace SYMBOLIC_BIND
with SYMBOL_REFERENCES_LOCAL when checking R_386_GOTOFF against
protected data symbol when building shared library.
* elf64-x86-64.c (elf_x86_64_relocate_section): Check
R_X86_64_GOTOFF64 against undefined symbol and replace
SYMBOLIC_BIND with SYMBOL_REFERENCES_LOCAL when checking
R_X86_64_GOTOFF64 against protected data symbol when building
shared library.
ld/testsuite/
* ld-i386/i386.exp: Run protected6a.
* ld-i386/protected6.d: Renamed to ...
* ld-i386/protected6a.d: This.
* ld-x86-64/hidden4.d: New file.
* ld-x86-64/hidden4.s: Likewise.
* ld-x86-64/hidden5.d: Likewise.
* ld-x86-64/hidden5.s: Likewise.
* ld-x86-64/protected6.d: Renamed to ...
* ld-x86-64/protected6a.d: This.
* ld-x86-64/x86-64.exp: Run hidden4, hidden5, protected6a,
protected7a and protected7b.
R_386_GOTOFF/R_X86_64_GOTOFF64 relocation shouldn't be used against
protected data symbol on x86 since with copy relocation, address of
protected data defined in the shared library may be external.
This patch will break building shared libraries with protected data
symbols using GCCs older than GCC 5 without the bug fix for
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65248
GCC backport request should be made in the GCC bug report above.
bfd/
PR ld/pr17709
* elf32-i386.c (elf_i386_relocate_section): Also check R_386_GOTOFF
against protected data symbol when building shared library.
* elf64-x86-64.c (elf_x86_64_relocate_section): Also check
R_X86_64_GOTOFF64 against protected data symbol when building
shared library.
ld/testsuite/
PR ld/pr17709
* ld-i386/protected6.d: New file.
* ld-i386/protected6.s: Likewise.
* ld-x86-64/protected6.d: Likewise.
* ld-x86-64/protected6.s: Likewise.
* ld-x86-64/protected7.d: Likewise.
* ld-x86-64/protected7.s: Likewise.
* ld-x86-64/protected7a.d: Likewise.
* ld-x86-64/protected7b.d: Likewise.
Re-apply: commit ca3fe95e46
With copy relocation, address of protected data defined in the shared
library may be external. This patch adds extern_protected_data and
changes _bfd_elf_symbol_refs_local_p to return false for protected data
if extern_protected_data is true.
This patch will break building shared libraries with protected data
symbols using GCCs older than GCC 5 without the bug fix for
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65248
GCC backport request should be made in the GCC bug report above.
To get correct run-time behavior on Linux, glibc 2.22 or above are
required, which have the bug fix for
https://sourceware.org/bugzilla/show_bug.cgi?id=17711
Backports for glibc 2.21, 2.20, 2.19 and 2.18 are on hjl/pr17711/2.21,
hjl/pr17711/2.20, hjl/pr17711/2.19 and hjl/pr17711/2.18 branches,
respectively, at
https://sourceware.org/git/?p=glibc.git;a=summary
bfd/
PR ld/pr15228
PR ld/pr17709
* elf-bfd.h (elf_backend_data): Add extern_protected_data.
* elf32-i386.c (elf_backend_extern_protected_data): New.
Defined to 1.
* elf64-x86-64.c (elf_backend_extern_protected_data): Likewise.
* elflink.c (_bfd_elf_adjust_dynamic_copy): Don't error on
copy relocs against protected symbols if extern_protected_data
is true.
(_bfd_elf_symbol_refs_local_p): Don't return true on protected
non-function symbols if extern_protected_data is true.
* elfxx-target.h (elf_backend_extern_protected_data): New.
Default to 0.
(elfNN_bed): Initialize extern_protected_data with
elf_backend_extern_protected_data.
ld/testsuite/
PR ld/pr15228
PR ld/pr17709
* ld-i386/i386.exp (i386tests): Add a test for PR ld/17709.
* ld-i386/pr17709-nacl.rd: New file.
* ld-i386/pr17709.rd: Likewise.
* ld-i386/pr17709a.s: Likewise.
* ld-i386/pr17709b.s: Likewise.
* ld-i386/protected3.d: Updated.
* ld-i386/protected3.s: Likewise.
* ld-x86-64/pr17709-nacl.rd: New file.
* ld-x86-64/pr17709.rd: Likewise.
* ld-x86-64/pr17709a.s: Likewise.
* ld-x86-64/pr17709b.s: Likewise.
* ld-x86-64/protected3.d: Updated.
* ld-x86-64/protected3.s: Likewise.
* ld-x86-64/x86-64.exp (x86_64tests): Add a test for PR ld/17709.
Compressed section contents should be bfd_alloced to avoid memory leak.
This patch replaces bfd_malloc and free with bfd_alloc and bfd_release
on compressed buffer in bfd_compress_section_contents. There is still
a very small memory leak when compressed section isn't smaller.
* compress.c (bfd_compress_section_contents): Replace bfd_malloc
and free with bfd_alloc and bfd_release on compressed buffer.
Release buffer if compressed section isn't smaller.
We can't check if input is ELF in copy_file since some targets may only
set xvec after bfd_check_format_matches is called. This patch moves
this check to copy_object.
bfd/
* elfxx-target.h (TARGET_BIG_SYM): Add BFD_COMPRESS_GABI to
object_flags.
(TARGET_LITTLE_SYM): Likewise.
binutils/
* objcopy.c (do_debug_sections): Use bit patterns.
(copy_object): Return FALSE for compress_zlib, compress_gnu_zlib
and compress_gabi_zlib on non-ELF input.
(copy_file): Don't check non-ELF input here.
text_action_add uses linear list search to order text actions list by
action VMA. The list is used at the first relaxation pass, when it's not
fixed yet.
Replace the list with splay tree from libiberty.
Original profile:
% time self children called name
-----------------------------------------
0.00 0.00 14/158225 compute_text_actions
3.62 0.00 25211/158225 remove_dead_literal
8.42 0.00 58645/158225 coalesce_shared_literal
10.68 0.00 74355/158225 text_action_add_proposed
38.8 22.73 0.00 158225 text_action_add
0.00 0.00 144527/293246 bfd_zmalloc
-----------------------------------------
Same data, after optimization:
% time self children called name
-----------------------------------------
0.00 0.00 14/158225 compute_text_actions
0.00 0.00 25211/158225 remove_dead_literal
0.00 0.01 58645/158225 coalesce_shared_literal
0.00 0.01 74355/158225 text_action_add_proposed
0.1 0.00 0.02 158225 text_action_add
0.01 0.00 144527/144527 splay_tree_insert
0.00 0.00 144527/195130 splay_tree_lookup
0.00 0.00 144527/293246 bfd_zmalloc
-----------------------------------------
2015-04-03 Max Filippov <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (splay-tree.h): include header.
(text_action_struct): drop next pointer.
(text_action_list_struct): drop head pointer, add count and
tree fields.
(find_fill_action): instead of linear search in text_action_list
search in the tree.
(text_action_compare, action_first, action_next): new functions.
(text_action_add, text_action_add_literal): instead of linear
search and insertion insert new node into the tree.
(removed_by_actions): pass additional parameter: action_list,
use it to traverse the tree.
(offset_with_removed_text): pass additional action_list parameter
to removed_by_actions.
(map_action_fn_context): new typedef.
(map_action_fn_context_struct): new structure.
(map_action_fn): new function.
(map_removal_by_action): use splay_tree_foreach to build map.
(find_insn_action): replace linear search in text_action_list
with series of splay_tree_lookups.
(print_action, print_action_list_fn): new functions.
(print_action_list): use splay_tree_foreach.
(init_xtensa_relax_info): drop action_list.head initialization.
Initialize the tree.
(compute_text_actions): use non-zero action_list_count instead of
non-NULL action list.
(xlate_map_context): new typedef.
(xlate_map_context_struct): new structure.
(xlate_map_fn): new function.
(build_xlate_map): use splay_tree_foreach to build map.
(action_remove_bytes_fn): new function.
(relax_section): use zero action_list_count instead of NULL
action list. Use splay_tree_foreach to count final section size.
Drop unused variable 'removed'.
find_removed_literal uses linear search to find removed literal by its
VMA. The list of literals is fixed at that point, build an ordered index
array and use binary search instead.
Original profile:
% time self children called name
-----------------------------------------
56.72 0.00 297578/669392 translate_reloc
70.86 0.00 371814/669392 relax_section
67.9 127.58 0.00 669392 find_removed_literal
-----------------------------------------
Same data, after optimization:
% time self children called name
-----------------------------------------
0.00 0.00 297578/669392 translate_reloc
0.00 0.00 371814/669392 relax_section
0.0 0.00 0.00 669392 find_removed_literal
0.00 0.00 23838/23838 map_removed_literal
-----------------------------------------
2015-04-03 Max Filippov <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (removed_literal_map_entry): new typedef.
(removed_literal_map_entry_struct): new structure.
(removed_literal_list_struct): add new fields: n_map and map.
(map_removed_literal, removed_literal_compare): new functions.
(find_removed_literal): build index array for literals ordered
by VMA, use binary search to find removed literal.
The function removed_by_actions iterates through text actions to
calculate an offset applied by text actions to a given VMA. Although it
has a parameter p_start_action that allows for incremental offset
calculation, in many places it's used with p_start_action explicitly set
to the first action. After the first relaxation pass when the list of
text actions is finalized, an array of offsets sorted by VMA may be used
to speed up this function.
Original profile:
% time self children called name
-----------------------------------------
0.35 0.00 33872/4808961 relax_section_symbols
3.32 0.00 326022/4808961 relax_property_section
12.83 0.00 1259379/4808961 offset_with_removed_text
32.50 0.00 3189688/4808961 translate_reloc
71.5 49.00 0.00 4808961 removed_by_actions
-----------------------------------------
Same data, after optimization:
% time self children called name
-----------------------------------------
0.00 0.00 33872/4808537 relax_section_symbols
0.01 0.00 326022/4808537 relax_property_section
0.05 0.00 1258955/4808537 offset_with_removed_text_map
0.13 0.00 3189688/4808537 translate_reloc
1.0 0.20 0.00 4808537 removed_by_actions_map
0.00 0.00 120/120 map_removal_by_action
-----------------------------------------
2015-04-01 Max Filippov <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (removal_by_action_entry_struct,
removal_by_action_map_struct): new structures.
(removal_by_action_entry, removal_by_action_map): new typedefs.
(text_action_list_struct): add new field: map.
(map_removal_by_action, removed_by_actions_map,
offset_with_removed_text_map): new functions.
(relax_section): replace offset_with_removed_text with
offset_with_removed_text_map.
(translate_reloc, relax_property_section, relax_section_symbols):
replace removed_by_actions with removed_by_actions_map.
The original check_section_ebb_pcrels_fit algorithm checks that text
actions proposed for current EBB are OK for every relocation in a
section. There's no need to check every relocation, because text actions
for EBB can only change size of that EBB, thus only affecting
relocations that in any way cross that EBB. In addition EBBs are
iterated in ascending order of their VMA, making it easier to track
relevant relocations.
Introduce a structure that can track relocations that cross the range of
VMAs of EBB and use it to only check relocations relevant to current EBB
in check_section_ebb_pcrels_fit.
It takes O(N log N) operations to build it and O(N) operations to move
current EBB VMA window through its entire range, where N is the number
of relocations in a section. The resulting complexity of
compute_text_actions is thus reduced from O(N^2) to O(N log N + N * M),
where M is the average number of relocations crossing each EBB.
Original profile:
% time self children called name
-----------------------------------------
44.26 71.53 6429/6429 compute_text_actions
50.2 44.26 71.53 6429 check_section_ebb_pcrels_fit
1.16 20.12 347506666/347576152 pcrel_reloc_fits
2.95 16.52 347506666/348104944 get_relocation_opnd
2.01 9.74 347575100/361252208 r_reloc_init
0.55 7.53 347575100/363381467 r_reloc_get_section
5.76 0.02 695013332/695013332 xlate_offset_with_removed_text
0.68 3.89 347575100/363483827 bfd_octets_per_byte
0.32 0.00 347506666/349910253 is_alt_relocation
0.18 0.11 6391/6391 build_xlate_map
0.00 0.00 6429/19417168 get_xtensa_relax_info
0.00 0.00 6391/6391 free_xlate_map
-----------------------------------------
Same data, after optimization:
% time self children called name
-----------------------------------------
2.56 3.08 6429/6429 compute_text_actions
8.2 2.56 3.08 6429 check_section_ebb_pcrels_fit
0.08 0.91 17721075/17790561 pcrel_reloc_fits
0.17 0.47 17721075/31685977 r_reloc_init
0.43 0.00 35442150/35442150 xlate_offset_with_removed_text
0.02 0.37 17721075/33815236 r_reloc_get_section
0.22 0.11 6391/6391 build_xlate_map
0.05 0.22 17721075/33917596 bfd_octets_per_byte
0.03 0.00 17721075/20405299 is_alt_relocation
0.01 0.00 6429/6429 reloc_range_list_update_range
0.00 0.00 6429/19417168 get_xtensa_relax_info
0.00 0.00 6391/6391 free_xlate_map
-----------------------------------------
2015-04-01 Max Filippov <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (reloc_range_list, reloc_range_list_entry,
reloc_range): new typedef.
(reloc_range_list_struct, reloc_range_list_entry_struct,
reloc_range_struct): new structures.
(reloc_range_compare, build_reloc_ranges,
reloc_range_list_append, reloc_range_list_remove,
reloc_range_list_update_range, free_reloc_range_list): new
functions.
(compute_text_actions): precompute relocation opcodes before the
loop. Add relevant_relocs variable, initialize it before the
loop, pass it to the check_section_ebb_pcrels_fit.
(check_section_ebb_pcrels_fit): add new parameter:
relevant_relocs. Update address range in the relevant_relocs if
it's non-NULL and iterate only over relevant relocations.
bfd * elf32-rx.c (describe_flags): Report the settings of the string
insn using bits.
(rx_elf_merge_private_bfd_data): Handle merging of the string insn
using bits.
bin * readelf.c (get_machine_flags): Report the setting of the string
insn using bits.
gas * config/tc-rx.c (enum options): Add OPTION_DISALLOW_STRING_INSNS.
(md_longopts): Add -mno-allow-string-insns.
(md_parse_option): Handle -mno-allow-string-insns.
(md_show_usage): Mention -mno-allow-string-insns.
(rx_note_string_insn_use): New function. Produces an error
message if a string insn is used when it is not allowed.
* config/rx-parse.y (SCMPU): Call rx_note_string_insn_use.
(SMOVU, SMOVB, SMOVF, SUNTIL, SWHILE, RMPA): Likewise.
* config/rx-defs.h (rx_note_string_insn_use): Prototype.
* doc/c-rx.texi: Document -mno-allow-string-insns.
elf * rx.h (E_FLAG_RX_SINSNS_SET): New bit in e_flags field.
(E_FLAG_RX_SINSNS_YES): Likewise.
(E_FLAG_RX_SINSNS_MASK): New define.
H.J. Lu
Jiong Wang
Nick Clifton
GDB Administrator
Alan Modra
John Baldwin
Max Filippov